Convolutional Neural Network Training with 1x1 Spatial Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for training fully convolutional neural networks for image recognition processes require a large amount of manually segmented training data, leading to increased workload and potential reduction in precision.
Innovation Solution
The proposed solution involves using a convolutional neural network that excludes the upsampling process and uses training data with a 1×1 spatial dimension, reducing the workload required to produce training data while maintaining precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the number of training data items is increased to improve image recognition precision, then the precision of the image recognition process is improved, but the workload required for manual segmentation of training data increases
Solution Approach 1:
The system automatically generates training data by capturing images and automatically determining their classifications, eliminating the need for manual segmentation. The processing device performs automated image classification using the trained neural network, and the classified images are stored as training data for future use, creating a self-sustaining system that reduces manual workload while maintaining data quality
Solution Approach 2:
The system performs preliminary classification of captured images before they are used as training data. By pre-classifying images during the data collection phase, the system prepares training data in advance with correct labels, reducing the need for subsequent manual segmentation work and enabling faster training iterations
2Measurement precision
If the number of training data items is increased to improve image recognition precision, then the precision of the image recognition process is improved, but the time required for training increases
Solution Approach 1:
The system continuously captures images and continuously adds classified images to the training data set. This continuous data accumulation allows the training process to benefit from progressively larger data sets without requiring periodic batch processing, enabling more efficient training iterations as the data set grows over time
Solution Approach 2:
The system creates copies of captured images and stores them in the training data set with automatic classifications. These copied images serve as additional training samples, increasing the data set size and improving recognition precision without requiring original physical samples or extensive manual processing of each image
Data Source
Figure 1A~1C
Figure 2
Figure 3
AI summary
An input unit 132 receives an input of an image subject to a process. A processing unit subjects the image input to the input unit 132 to a process of a convolutional neural network in which a fully connected layer is excluded. The convolutional neural network in the processing unit 114 includes a convolutional layer and a pooling layer. An output unit outputs a result of the process in the processing unit 114. A filter of the convolutional layer in the convolutional neural network in the processing unit 114 is trained to learn the result of the process having a 1×1 spatial dimension.